NEWS: quick list of everything changed since 1.11.0

will need lots of editing before release
This commit is contained in:
Brian Warner 2016-10-19 12:00:04 -07:00
parent b96122a8ff
commit 71c2b20507
11 changed files with 167 additions and 109 deletions

167
NEWS.rst
View File

@ -4,6 +4,173 @@
User-Visible Changes in Tahoe-LAFS User-Visible Changes in Tahoe-LAFS
================================== ==================================
Release 1.12.0 (?)
''''''''''''''''''
#1720
2367.removal:
The little-used "manhole" debugging feature has been removed. This allowed
you to SSH or Telnet "into" a Tahoe node, providing an interactive
Read-Eval-Print-Loop (REPL) that executed inside the context of the running
process. The SSH authentication code used a deprecated feature of Twisted,
this code had no unit-test coverage, and I haven't personally used it in at
least 6 years (despite writing it in the first place). Time to go.
2491.config:
* tub.port is now an Endpoint server specification string (which is pretty
much just like a strports string, but can be extended by plugins). It now
rejects "tcp:0" and "0". The tahoe.cfg value overrides anything stored on
disk (in client.port). This should have no effect on most nodes, which do
not set tub.port in tahoe.cfg, and wrote an allocated port number to
client.port the first time they launched. Folks who want to listen on a
specific port number typically set tub.port to "tcp:12345" or "12345", not
"0".
* The "portnumfile" (e.g. NODEDIR/client.port) is written as soon as the port
is allocated, before the tub is created, and only if "tub.port" was empty.
The old code wrote to it unconditionally, and after Tub startup. So if the
user allows NODEDIR/client.port to be written, then later modifies
tahoe.cfg to set "tub.port" to a different value, this difference will
persist (and the node will honor tahoe.cfg "tub.port" exclusively).
* We now encourage static allocation of tub.port, and pre-configuration of
the node's externally-reachable IP address or hostname (by setting
tub.location). Automatic IP-address detection is deprecated. Automatic port
allocation is merely discouraged. Eventually both will be managed by "tahoe
create-node", but for now we recommend users edit their tahoe.cfg after
node creation and before first launch.
* "tahoe start" now creates the Tub, and all primary software components,
before the child process daemonizes. Many configuration errors which would
previously have been reported in a logfile (after node startup), will now
be signalled immediately, via stderr. In these cases, the "tahoe start"
process will exit with a non-zero return code.
2759.docs:
Tahoe now uses a separate Foolscap tub for each outbound storage server
connection. This has two benefits:
* a slight privacy improvement: storage servers can no longer compare client
TubIDs to confirm/deny that two clients are the same (but note there are
other reliable signals: timing correlations, interest in the same shares,
future Accounting identifiers)
* this enables future per-server connection options, like using Tor for some
servers but direct TCP connections for others (#517).
and a few drawbacks:
* It causes a small performance hit to generate new TLS keys (2048-bit RSA)
for each connection. On a modern computer, this adds 75ms per server.
* It breaks a NAT-bypass trick which enabled storage servers to run behind
NAT boxes, which was only useful if all the *clients* of the storage server
had public IP addresses, and those clients were also configured as servers.
The trick was to configure the NAT-bound server as a client too: its
outbound connections to the "servers" would be used in the opposite
direction to provide storgae service to the clients (Foolscap doesn't care
who initiated the connection, as long as both ends have the right TLS
keys). We decided that this trick is not sufficiently general to preserve.
* Server logs that record a TubID are no longer so easy to use: until
Accounting (#666) lands and a client-id is used for log messages, it will
be difficult to identify exactly which client the log is referencing.
2773.docs:
The "stats-gatherer", an operation-helper service used to collect runtime
statistics from a fleet of Tahoe storage servers, must now be assigned a
hostname, or location+port pair, at creation time. The "tahoe
create-stats-gatherer" command now requires either "--hostname=", or both
"--location=" and "--port".
Previously, "tahoe create-stats-gatherer NODEDIR" would attempt to guess its
location by running something like /sbin/ifconfig to collect local IP
addresses. While this works if the host has a public IP address (or at least
lives in the same LAN as the storage servers it monitors), most sysadmins
would prefer the FURL be created with a real hostname.
To keep your old stats-gatherers working, with their original FURL, you must
determine a suitable --location and --port, and write their values into
NODEDIR/location and NODEDIR/port, respectively. Or you could simply rebuild
it by re-running "tahoe create-stats-gatherer" with the new arguments.
See docs/stats.rst for details.
2776.change:
Tahoe's testing-only dependencies can now be installed by asking for the
[test] extra, so if you want to set up a virtualenv for testing, use "pip
install -e .[test]" instead just of "pip install -e ." . At the moment this
only includes "coverage" and "pyflakes", but in the future it might include
"mock" and other utility libraries.
2781.packaging:
Tahoe now requires Twisted >= 16.1.0, so ensure that unit tests do not fail
because of uncancelled timers left running by HostnameEndpoint.
2783.docs:
The "key-generator" node type has been removed. This was a standalone process
that maintained a queue of RSA keys. Clients could offload the key-generation
work by adding "key_generator.furl=" in their tahoe.cfg files, to create
mutable files and directories faster. This seemed important back in 2006, but
these days computers are faster and RSA key generation only takes about 90ms.
This removes the "tahoe create-key-generator" command. Any
"key_generator.furl" settings in tahoe.cfg will log a warning and otherwise
ignored. Attempts to "tahoe start" a previously-generated key-generator node
will result in an error.
2794.change:
The little-used "control port" now uses a separate (ephemeral) Tub. This
means the FURL changes each time the node is restarted, and it only listens
on the loopback (127.0.0.1) interface, on a random port. As the control port
is only used by some automated tests (check_memory, check_speed), this
shouldn't affect anyone.
The slightly-more-used "log port" now also uses a separate (ephemeral) Tub,
with the same consequences. The lack of a stable (and externally-reachable)
logport.furl means it is no longer possible to use `flogtool tail FURL`
against a distant Tahoe server, however `flogtool tail
.../nodedir/private/logport.furl` still works just fine (and is the more
common use case anyways). We might bring back the ability to configure the
port and location of the logport in the future, if there is sufficient
demand, but for now it seems better to avoid the complexity.
PR242.docs:
The "stats gatherer" (created with 'tahoe create-stats-gatherer') now updates
a JSON file named "stats.json"; previously it used Pickle and "stats.pickle".
The munin plugins in misc/operations_helpers/munin/ have been updated to
match, and must be re-installed and re-configured if you use munin.
more:
* lots of tox environment cleanups, moving most tooling from Makefile to tox,
including code-checks, deprecation warnings, code-coverage
* 2774 add internal progress API (for magic-folder)
* 1903 document "/file/" as deprecated, replace with "/named/"
* 1449 don't start magic-folder until "connected enough"
* 2491 start Tub synchronously, improves error delivery at startup
* docs are now hosted at tahoe-lafs.readsthedocs.io (not .org)
* client uses separate Tub per storage server (privacy improvement)
* 2788 introducer cache, yaml file, loaded at startup until real introducer
arrives: tolerates introducer failure better. static servers can be
configured by adding a servers.yaml
* 2784 remove v1 introducer (sigs now required)
* BIG replace drop-upload with magic-folders
* 2754 remove _appname.py
* 1363 use pubkey as serverid, not tubid. affects new-helper vs old-uploader
* 517: connection handlers: socks/tor/i2p, tahoe.cfg [connections]
* 2810 disable foolscap gifts
* 2816 empty tub.port means don't listen at all
* 1010 tahoe.cfg reveal-ip-address = false, create-client --hide-ip
* 2818 show connection hints on welcome page
* 1942 remove google-based timing chart on mapupdate page (privacy)
* 2809 reject unrecognized tahoe.cfg options at startup
* 2735 CLI remove 'debug trial' and 'debug repl'
* 2773 create-node/create-introducer --location/--port --hostname --listen
* 68 introducers.yaml, multiple introducers, no introducers
* 2827 explain ipv6
* 867 tub.port takes multiple endpoints (comma-separated)
* 2490 --listen=tor
* fix IP-autodetect on RHEL (for tests)
Release 1.11.0 (30-Mar-2016) Release 1.11.0 (30-Mar-2016)
'''''''''''''''''''''''''''' ''''''''''''''''''''''''''''

View File

View File

@ -1,6 +0,0 @@
The little-used "manhole" debugging feature has been removed. This allowed
you to SSH or Telnet "into" a Tahoe node, providing an interactive
Read-Eval-Print-Loop (REPL) that executed inside the context of the running
process. The SSH authentication code used a deprecated feature of Twisted,
this code had no unit-test coverage, and I haven't personally used it in at
least 6 years (despite writing it in the first place). Time to go.

View File

@ -1,25 +0,0 @@
* tub.port is now an Endpoint server specification string (which is pretty
much just like a strports string, but can be extended by plugins). It now
rejects "tcp:0" and "0". The tahoe.cfg value overrides anything stored on
disk (in client.port). This should have no effect on most nodes, which do
not set tub.port in tahoe.cfg, and wrote an allocated port number to
client.port the first time they launched. Folks who want to listen on a
specific port number typically set tub.port to "tcp:12345" or "12345", not
"0".
* The "portnumfile" (e.g. NODEDIR/client.port) is written as soon as the port
is allocated, before the tub is created, and only if "tub.port" was empty.
The old code wrote to it unconditionally, and after Tub startup. So if the
user allows NODEDIR/client.port to be written, then later modifies
tahoe.cfg to set "tub.port" to a different value, this difference will
persist (and the node will honor tahoe.cfg "tub.port" exclusively).
* We now encourage static allocation of tub.port, and pre-configuration of
the node's externally-reachable IP address or hostname (by setting
tub.location). Automatic IP-address detection is deprecated. Automatic port
allocation is merely discouraged. Eventually both will be managed by "tahoe
create-node", but for now we recommend users edit their tahoe.cfg after
node creation and before first launch.
* "tahoe start" now creates the Tub, and all primary software components,
before the child process daemonizes. Many configuration errors which would
previously have been reported in a logfile (after node startup), will now
be signalled immediately, via stderr. In these cases, the "tahoe start"
process will exit with a non-zero return code.

View File

@ -1,25 +0,0 @@
Tahoe now uses a separate Foolscap tub for each outbound storage server
connection. This has two benefits:
* a slight privacy improvement: storage servers can no longer compare client
TubIDs to confirm/deny that two clients are the same (but note there are
other reliable signals: timing correlations, interest in the same shares,
future Accounting identifiers)
* this enables future per-server connection options, like using Tor for some
servers but direct TCP connections for others (#517).
and a few drawbacks:
* It causes a small performance hit to generate new TLS keys (2048-bit RSA)
for each connection. On a modern computer, this adds 75ms per server.
* It breaks a NAT-bypass trick which enabled storage servers to run behind
NAT boxes, which was only useful if all the *clients* of the storage server
had public IP addresses, and those clients were also configured as servers.
The trick was to configure the NAT-bound server as a client too: its
outbound connections to the "servers" would be used in the opposite
direction to provide storgae service to the clients (Foolscap doesn't care
who initiated the connection, as long as both ends have the right TLS
keys). We decided that this trick is not sufficiently general to preserve.
* Server logs that record a TubID are no longer so easy to use: until
Accounting (#666) lands and a client-id is used for log messages, it will
be difficult to identify exactly which client the log is referencing.

View File

@ -1,18 +0,0 @@
The "stats-gatherer", an operation-helper service used to collect runtime
statistics from a fleet of Tahoe storage servers, must now be assigned a
hostname, or location+port pair, at creation time. The "tahoe
create-stats-gatherer" command now requires either "--hostname=", or both
"--location=" and "--port".
Previously, "tahoe create-stats-gatherer NODEDIR" would attempt to guess its
location by running something like /sbin/ifconfig to collect local IP
addresses. While this works if the host has a public IP address (or at least
lives in the same LAN as the storage servers it monitors), most sysadmins
would prefer the FURL be created with a real hostname.
To keep your old stats-gatherers working, with their original FURL, you must
determine a suitable --location and --port, and write their values into
NODEDIR/location and NODEDIR/port, respectively. Or you could simply rebuild
it by re-running "tahoe create-stats-gatherer" with the new arguments.
See docs/stats.rst for details.

View File

@ -1,5 +0,0 @@
Tahoe's testing-only dependencies can now be installed by asking for the
[test] extra, so if you want to set up a virtualenv for testing, use "pip
install -e .[test]" instead just of "pip install -e ." . At the moment this
only includes "coverage" and "pyflakes", but in the future it might include
"mock" and other utility libraries.

View File

@ -1,2 +0,0 @@
Tahoe now requires Twisted >= 16.1.0, so ensure that unit tests do not fail
because of uncancelled timers left running by HostnameEndpoint.

View File

@ -1,10 +0,0 @@
The "key-generator" node type has been removed. This was a standalone process
that maintained a queue of RSA keys. Clients could offload the key-generation
work by adding "key_generator.furl=" in their tahoe.cfg files, to create
mutable files and directories faster. This seemed important back in 2006, but
these days computers are faster and RSA key generation only takes about 90ms.
This removes the "tahoe create-key-generator" command. Any
"key_generator.furl" settings in tahoe.cfg will log a warning and otherwise
ignored. Attempts to "tahoe start" a previously-generated key-generator node
will result in an error.

View File

@ -1,14 +0,0 @@
The little-used "control port" now uses a separate (ephemeral) Tub. This
means the FURL changes each time the node is restarted, and it only listens
on the loopback (127.0.0.1) interface, on a random port. As the control port
is only used by some automated tests (check_memory, check_speed), this
shouldn't affect anyone.
The slightly-more-used "log port" now also uses a separate (ephemeral) Tub,
with the same consequences. The lack of a stable (and externally-reachable)
logport.furl means it is no longer possible to use `flogtool tail FURL`
against a distant Tahoe server, however `flogtool tail
.../nodedir/private/logport.furl` still works just fine (and is the more
common use case anyways). We might bring back the ability to configure the
port and location of the logport in the future, if there is sufficient
demand, but for now it seems better to avoid the complexity.

View File

@ -1,4 +0,0 @@
The "stats gatherer" (created with 'tahoe create-stats-gatherer') now updates
a JSON file named "stats.json"; previously it used Pickle and "stats.pickle".
The munin plugins in misc/operations_helpers/munin/ have been updated to
match, and must be re-installed and re-configured if you use munin.